I have a more general question about supervised classification, I hope this post is not in the wrong category.
To apply a classifier from one image to another requires very carefully selected training areas. If depends a bit on the classifier, but random forest, for example, uses lots of fine threholds to classifiy the training input. In most cases, application of the classifier to other images doesn’t work very good. Also, for example, because Sentinel-1 and ENVISAT have different pixel sizes which represent different areas of backscatter.
Thanks for replying ABraun.
Concerning pixel sizes, a resampling operation would solve this issue (I think). We are going to use SNAP in the first place. Would you suggest any specific classification method for this application? I am considering KNN which is pretty straight forward.
The problem is no that they have different pixel sizes but that these pixels represent different areas in terms of backscattering. So even calibrated images are not directly comparable if the pixel sizes are too different. But it is surely worth a try.
KNN is probably a good choice, yes.
Thanx again ABraun. I will probably give it a try. Best, Efi